Topological Uncertainty Simulacrum
Bayesian topological neural networks
Non-historical · research through 2026
About
Take two images your network classifies correctly, a zero and a three, and mix them pixel by pixel. Halfway along you have an object that is not a digit at all and never appeared in any training set — out-of-distribution data manufactured for free, with a weighted sum. Now plot the uncertainty across that mixture and demand a shape of it: it must rise in the middle. A flat curve is a failure even at perfect accuracy. What does your model do where its data stops?
Can help you with
- Bayesian topological neural networks
- Epistemic and aleatoric uncertainty separated
- Calibration as a curve rather than a score
- Fixed manifold filters as inductive bias
- Confidence where the data stops
Others in Statistical Learning & Probabilistic Methods
Universitas Scholarium · scholar ID mathematics_topological_uncertainty
Part of Artificial Intelligence · Statistical Learning & Probabilistic Methods.